arXiv subjects

Chaozheng Wang

Publications and source records attributed to Chaozheng Wang.

At least 19 recordsLinked to original sources

arXiv2026

LongPuzzleBench: Evaluating GUI Agents on Long-Horizon Visual Puzzles

GUI agents need long-horizon visual reasoning: they must interpret a changing interface while keeping a multi-step plan viable as earlier actions constrain later ones. Existing benchmarks evaluate grounding, computer use, and game play, but rarely test whether agents stay coherent across long chains of coupled decisions. Long-horizon visual puzzles expose this capability directly: a legal move that looks like progress can make the puzzle unsolvable, and the loss shows only several moves later. We introduce LongPuzzleBench, 114 levels in six puzzle games played through native GUI actions, where one objective can take a human over a thousand actions on persistent boards and dead ends go unannounced. With Native GUI Actions alone, the strongest agents solve most objectives, but success falls sharply on harder, longer boards: seven of ten general-purpose agents solve nothing harder than Medium, and none completes Bolt Unscrew Hard, which a human solves along with every other objective. Code Execution CUA does not close this gap, and its scores mix visual solving with algorithmic search. Controlled diagnostics trace these failures to one limitation that neither rules, state hints, nor failure memory removes: agents judge each move by the visible progress it makes, not by the future options it leaves.

cs.CL↗

arXiv2026

AgenticDB: Self-Evolving Reconfiguration Framework for Database Workloads

Configuration tuning is critical to database performance but remains difficult in real deployments. Despite notable advances, prior methods still leave substantial performance potential unexplored, suffer from low tuning efficiency, and provide limited support for configuration validation and failure recovery. To address these limitations, we propose AgenticDB, a self-evolving agentic framework for database workload reconfiguration. AgenticDB uses a large language model (LLM)-based DBA Planner to jointly reconfigure database knobs and operating system (OS) parameters through two key mechanisms. First, context-grounded bottleneck diagnosis uses workload characteristics, configuration state, and observed runtime behavior to identify the current performance bottleneck and recommend targeted database management system (DBMS)/OS reconfiguration actions. Second, closed-loop context evolution uses observed performance and runtime-state changes as feedback to update the bottleneck diagnosis, guide subsequent decisions, and terminate the reconfiguration loop when performance plateaus. It also consolidates accumulated reconfiguration experience for reuse on workloads with similar characteristics. Beyond these two mechanisms, AgenticDB improves reliability by validating each proposed configuration before applying it and automatically recovering from failures. We evaluate AgenticDB on MySQL and PostgreSQL using YCSB, Sysbench, and TPC-H. Compared with SOTA methods, AgenticDB outperforms the best-performing baseline by 118.1% on average and reduces the total time-to-best across workloads by 22.6%. Further analyses show that validation and recovery improve reconfiguration reliability. The consolidated experience also helps AgenticDB reach high-performing configurations earlier on workloads with similar characteristics.

cs.DB↗

arXiv2026

Code Benchmarks Should Prioritize Rigor, Reliability, and Reproducibility

Code-related benchmarks play a critical role in evaluating large language models (LLMs), yet their quality fundamentally shapes how the community interprets model capabilities. In the past few years, awareness of benchmark quality has grown. Yet, after a decade-scale (2014-2025) survey over 672 code benchmarks, we observed a lag between growing awareness and actual practice. For example, in 2025 alone, the number of benchmarks that ignore code coverage when providing test cases nearly matches the total count accumulated across the previous ten years. In response, we take a clear position: Code benchmarks must prioritize rigor in benchmark construction, reliability in evaluation, and reproducibility in release. To operationalize this position, we introduce a code benchmark guideline HOW2BENCH with 55 checklists. Finally, our further human study also exposed that the current issues not only stem from the significant effort required, but also from a lack of awareness regarding their importance.

cs.SE↗

arXiv2026

SWE-MeM: Learning Adaptive Memory Management for Long-Horizon Coding Agents

Long-horizon software engineering agents often need to manage lengthy and noisy interaction histories under limited context budgets. Existing memory management methods typically rely on static compression workflows or impose rigid constraints on compression timing and granularity. Moreover, these approaches fail to jointly optimize memory management and issue resolution capabilities to improve performance while reducing token usage. We present SWE-MeM, a training framework for proactive and on-demand memory management in software engineering agents. SWE-MeM provides a flexible memory tool that lets agents decide when, what, and how to compress based on trajectory state, task progress, and remaining context budget. We train agents with synthesized proactive memory-management trajectories and Memory-aware GRPO, which jointly optimizes memory management and issue resolution through memory-aware trajectory splitting and step-level credit assignment. On SWE-Bench Verified, SWE-MeM achieves 43.4% and 60.2% resolve rate with 4B and 30B models, respectively, outperforming existing memory management baselines in both performance and efficiency.

cs.SE↗

arXiv2026

Unlocking Proactivity in Task-Oriented Dialogue

Proactive task-oriented dialogue (TOD), such as outbound sales, demands a persuasive agent that actively probes the user's concerns and steers the conversation toward acceptance within a bounded number of turns. Yet post-trained LLMs are inherently conservative, and reward-shaping RL (e.g., GRPO) struggles since it only re-weights what an already passive policy samples. We show that conditioning on the user's latent concerns unlocks proactive capability that no amount of sampling can undermine, establishing these concerns as a pivotal training-time signal. To operationalize this finding, we build the \textbf{Cognitive User Simulator}, which models each user as a stratified persona comprising observable external traits and hidden internal concerns. The simulator produces faithful and diverse interactions, while emitting per-turn state dynamics that track persuasion progress. We then introduce \textbf{Simulator-Induced Asymmetric-View Policy Optimization}, which converts the modeled concerns and the simulation state transition into complementary training objectives: (1) \emph{Asymmetric On-Policy Self-Distillation} that transfers concern-aware behavior from a privileged view of the same policy into its deployable, conversation-only view; and (2) \emph{State-Transition Policy Refinement} ...

cs.AI↗

arXiv2026

Dependency-Guided Repository-Level C-to-Rust Translation with Reinforcement Alignment

Automating C-to-Rust migration is critical for improving software security without sacrificing performance. Traditional rule-based methods struggle with diverse C idioms, often producing rigid and unidiomatic Rust code. Large Language Models (LLMs), trained on massive code corpora, offer a promising alternative by leveraging cross-language generalization to generate more idiomatic and maintainable Rust code. However, several challenges remain. First, existing LLM-based approaches fail to handle cross-file dependencies effectively, either ignoring them or including entire files as context, which limits accurate dependency modeling. Second, complex dependencies and structured inputs and outputs make it difficult to verify syntactic correctness and functional equivalence at the repository level. Third, the lack of large-scale C-Rust parallel data constrains model performance. We propose DepTrans, a framework that combines model capability enhancement with structured inference. DepTrans introduces Reinforcement-Aligned Syntax Training to improve generation quality through multi-task fine-tuning and feedback-driven reinforcement learning. It further applies Dependency-Guided Iterative Refinement to capture fine-grained cross-file dependencies and iteratively refine generated Rust code. We construct a dataset of 85k training samples and a benchmark of 145 repository-level instances. Experiments show that DepTrans achieves a 60.7 percent compilation success rate and 43.5 percent computational accuracy, outperforming the strongest baseline by 22.8 and 17.3 percentage points. It also successfully builds 7 of 15 industrial C projects, demonstrating its practical potential.

cs.SE↗

arXiv2026

Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning

On-policy self-distillation (OPSD) is an emerging LLM post-training paradigm in which the model serves as its own teacher: conditioned on privileged information such as a reference trace or hint, the same policy provides dense token-level supervision on its own rollouts. However, recent studies show that OPSD degrades complex reasoning by suppressing predictive uncertainty, which supports exploration and hypothesis revision. Our token-level analysis shows that this failure arises from applying a uniform direction of teacher supervision across tokens with different uncertainty levels: conformity to the privileged self-teacher suppresses exploration at high entropy, while deviation from the teacher degrades step accuracy at low entropy. Accordingly, we propose \textbf{Direction-Adaptive Self-Distillation} (\textbf{DASD}), which reframes privileged self-distillation from uniform teacher imitation into entropy-routed directional supervision: high-entropy tokens are pushed away from the privileged teacher to preserve exploration, while low-entropy tokens are pulled toward the teacher to stabilize step-level execution. Across six mathematical reasoning benchmarks, DASD achieves the best macro Avg@16 over strong RLVR and self-distillation baselines. Pass@$k$, reasoning-health, and generalization analyses show that these average gains come from preserving exploration without sacrificing step-level execution.

cs.LG↗

arXiv2026

SWE-Chain: Benchmarking Coding Agents on Chained Release-Level Package Upgrades

Coding agents powered by large language models are increasingly expected to perform realistic software maintenance tasks beyond isolated issue resolution. Existing benchmarks have shifted toward realistic software evolution, but they rarely capture continuous maintenance at the granularity of package releases, where changes are bundled, shipped, and inherited by subsequent versions. We present SWE-Chain, a benchmark for evaluating agents on chained release-level package upgrades, where each transition builds on the agent's prior codebase. To produce upgrade specifications, we design a divide-and-conquer synthesis pipeline that aligns release notes with code diffs for each version transition, ensuring the requirements are grounded in actual code changes, informative to agents, and feasible to implement. SWE-Chain contains 12 upgrade chains across 9 real Python packages, with 155 version transitions and 1,660 grounded upgrade requirements. Across nine frontier agent-model configurations, agents achieve an average of 44.8% resolving, 65.4% precision, and 50.2% F1 under the Build+Fix regime, with Claude-Opus-4.7 (Claude Code) leading at 60.8% resolving, 80.6% precision, and 68.5% F1. These results show that SWE-Chain is both feasible and discriminative, and reveal that current agents still struggle to make correct upgrades across chained package releases without breaking existing functionality.

cs.SE↗

arXiv2026

SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue

Large Language Models have demonstrated remarkable capabilities in open-domain dialogues. However, current methods exhibit suboptimal performance in service dialogues, as they rely on noisy, low-quality human conversation data. This limitation arises from data scarcity and the difficulty of simulating authentic, goal-oriented user behaviors. To address these issues, we propose SEAD (Self-Evolving Agent for Service Dialogue), a framework that enables agents to learn effective strategies without large-scale human annotations. SEAD decouples user modeling into two components: a Profile Controller that generates diverse user states to manage training curriculum, and a User Role-play Model that focuses on realistic role-playing. This design ensures the environment provides adaptive training scenarios rather than acting as an unfair adversary. Experiments demonstrate that SEAD significantly outperforms Open-source Foundation Models and Closed-source Commercial Models, improving task completion rate by 17.6% and dialogue efficiency by 11.1%. Code is available at: https://github.com/Da1yuqin/SEAD.

cs.CL↗

arXiv2026

Learning to Ask: When LLM Agents Meet Unclear Instruction

Equipped with the capability to call functions, modern large language models (LLMs) can leverage external tools for addressing a range of tasks unattainable through language skills alone. However, the effective execution of these tools relies heavily not just on the advanced capabilities of LLMs but also on precise user instructions, which often cannot be ensured in the real world. To evaluate the performance of LLMs tool-use under imperfect instructions, we meticulously examine the real-world instructions queried from users, analyze the error patterns, and build a challenging tool-use benchmark called Noisy ToolBench (NoisyToolBench). We find that due to the next-token prediction training objective, LLMs tend to arbitrarily generate the missed argument, which may lead to hallucinations and risks. To address this issue, we propose a novel framework, Ask-when-Needed (AwN), which prompts LLMs to ask questions to users whenever they encounter obstacles due to unclear instructions. Moreover, to reduce the manual labor involved in user-LLM interaction and assess LLMs performance in tool utilization from both accuracy and efficiency perspectives, we design an automated evaluation tool named ToolEvaluator. Our experiments demonstrate that the AwN significantly outperforms existing frameworks for tool learning in the NoisyToolBench. We will release all related code and datasets to support future research.

cs.CL↗

arXiv2026

Cascaded Code Editing: Large-Small Model Collaboration for Effective and Efficient Code Editing

Code editing constitutes a fundamental practice in software development, wherein developers modify existing codebases according to natural language requirements. Accurate code editing necessitates a comprehensive understanding of both the existing codebase and the modification requirements. Although large language models (LLMs) have demonstrated promising performance in code editing tasks, they suffer from substantial inefficiency by generating entire modified files that largely consist of unchanged code. While smaller models could potentially address this inefficiency, they typically lack the capacity to effectively comprehend long code contexts required for accurate editing. To ensure both effectiveness and efficiency, we propose to decompose code editing into a two-stage cascade: \textbf{edit sketch generation}, wherein a large model first produces concise sketches representing the requisite modifications (the more challenging phase), and \textbf{edit sketch application}, wherein a smaller model integrates these sketches into the original code to produce the final output edited code (the simpler phase). This cascaded design reduces the number of tokens generated by the large model, as the majority of the output is handled by the smaller, more efficient model, thereby enhancing overall efficiency. However, the effectiveness of this approach is constrained by current small models' limited capabilities in handling long-context scenarios and cross-file dependencies, which are essential for accurate sketch application in real-world codebases. To address these limitations and enhance smaller models' sketch application capabilities, ...

cs.SE↗

arXiv2026

On the Effectiveness of Context Compression for Repository-Level Tasks: An Empirical Investigation

Repository-level code intelligence tasks require large language models (LLMs) to process long, multi-file contexts. Such inputs introduce three challenges: crucial context can be obscured by noise, truncated due to limited windows, and increased inference latency. Context compression mitigates these risks by condensing inputs. While studied in NLP, its applicability to code tasks remains largely unexplored. We present the first systematic empirical study of context compression for repository-level code intelligence, organizing eight methods into three paradigms: discrete token sequences, continuous latent vectors, and visual tokens. We evaluate them on code completion and generation, measuring performance and efficiency. Results show context compression is effective: at 4x compression, continuous latent vector methods surpass full-context performance by up to 28.3% in BLEU score, indicating they filter noise rather than just truncating. On efficiency, all paradigms reduce inference cost. Both visual and text-based compression achieve up to 50% reduction in end-to-end latency at high ratios, approaching the cost of inference without repository context. These findings establish context compression as a viable approach and provide guidance for paradigm selection.

cs.SE↗

arXiv2026

SAGE: A Service Agent Graph-guided Evaluation Benchmark

The development of Large Language Models (LLMs) has catalyzed automation in customer service, yet benchmarking their performance remains challenging. Existing benchmarks predominantly rely on static paradigms and single-dimensional metrics, failing to account for diverse user behaviors or the strict adherence to structured Standard Operating Procedures (SOPs) required in real-world deployments. To bridge this gap, we propose SAGE (Service Agent Graph-guided Evaluation), a universal multi-agent benchmark for automated, dual-axis assessment. SAGE formalizes unstructured SOPs into Dynamic Dialogue Graphs, enabling precise verification of logical compliance and comprehensive path coverage. We introduce an Adversarial Intent Taxonomy and a modular Extension Mechanism, enabling low-cost deployment across domains and facilitating automated dialogue data synthesis. Evaluation is conducted via a framework where Judge Agents and a Rule Engine analyze interactions between User and Service Agents to generate deterministic ground truth. Extensive experiments on 27 LLMs across 6 industrial scenarios reveal a significant ``Execution Gap'' where models accurately classify intents but fail to derive correct subsequent actions. We also observe ``Empathy Resilience'', a phenomenon where models maintain polite conversational facades despite underlying logical failures under high adversarial intensity. Code and resources are available at https://anonymous.4open.science/r/SAGE-Bench-4CD3/.

cs.AI↗

arXiv2026

WARBENCH: A Comprehensive Benchmark for Evaluating LLMs in Military Decision-Making

Large Language Models are increasingly being considered for deployment in safety-critical military applications. However, current benchmarks suffer from structural blindspots that systematically overestimate model capabilities in real-world tactical scenarios. Existing frameworks typically ignore strict legal constraints based on International Humanitarian Law (IHL), omit edge computing limitations, lack robustness testing for fog of war, and inadequately evaluate explicit reasoning. To address these vulnerabilities, we present WARBENCH, a comprehensive evaluation framework establishing a foundational tactical baseline alongside four distinct stress testing dimensions. Through a large scale empirical evaluation of nine leading models on 136 high-fidelity historical scenarios, we reveal severe structural flaws. First, baseline tactical reasoning systematically collapses under complex terrain and high force asymmetry. Second, while state of the art closed source models maintain functional compliance, edge-optimized small models expose extreme operational risks with legal violation rates approaching 70 percent. Furthermore, models experience catastrophic performance degradation under 4-bit quantization and systematic information loss. Conversely, explicit reasoning mechanisms serve as highly effective structural safeguards against inadvertent violations. Ultimately, these findings demonstrate that current models remain fundamentally unready for autonomous deployment in high stakes tactical environments.

cs.CY↗

arXiv2026

Reinforcing Real-world Service Agents: Balancing Utility and Cost in Task-oriented Dialogue

The rapid evolution of Large Language Models (LLMs) has accelerated the transition from conversational chatbots to general agents. However, effectively balancing empathetic communication with budget-aware decision-making remains an open challenge. Since existing methods fail to capture these complex strategic trade-offs, we propose InteractCS-RL, a framework that reframes task-oriented dialogue as a multi-granularity reinforcement learning process. Specifically, we first establish a User-centric Interaction Framework to provide a high-fidelity training gym, enabling agents to dynamically explore diverse strategies with persona-driven users. Then, we introduce Cost-aware Multi-turn Policy Optimization (CMPO) with a hybrid advantage estimation strategy. By integrating generative process credits and employing a PID-Lagrangian cost controller, CMPO effectively guides the policy to explore Pareto boundary between user reward and global cost constraints. Extensive experiments on customized real business scenarios demonstrate that InteractCS-RL significantly outperform other baselines across three evaluation dimensions. Further evaluation on tool-agent-user interaction benchmarks verify InteractCS-RL robustness across diverse domains.

cs.CL↗

arXiv2026

REPAIR: Robust Editing via Progressive Adaptive Intervention and Reintegration

Post-training for large language models (LLMs) is constrained by the high cost of acquiring new knowledge or correcting errors and by the unintended side effects that frequently arise from retraining. To address these issues, we introduce REPAIR (Robust Editing via Progressive Adaptive Intervention and Reintegration), a lifelong editing framework designed to support precise and low-cost model updates while preserving non-target knowledge. REPAIR mitigates the instability and conflicts of large-scale sequential edits through a closed-loop feedback mechanism coupled with dynamic memory management. Furthermore, by incorporating frequent knowledge fusion and enforcing strong locality guards, REPAIR effectively addresses the shortcomings of traditional distribution-agnostic approaches that often overlook unintended ripple effects. Our experiments demonstrate that REPAIR boosts editing accuracy by 10%-30% across multiple model families and significantly reduces knowledge forgetting. This work introduces a robust framework for developing reliable, scalable, and continually evolving LLMs.

cs.CL↗

arXiv2025

SPVR: syntax-to-prompt vulnerability repair based on large language models

Purpose: In the field of vulnerability repair, previous research has leveraged pretrained models and LLM-based prompt engineering, among which LLM-based approaches show better generalizability and achieve the best performance. However, the LLM-based approaches generally regard vulnerability repair as a sequence-to-sequence task, and do not explicitly capture the syntax patterns for different vulnerability types, leading to limited accuracy. We aim to create a method that ensures the specificity of prompts targeting vulnerable code while also leveraging the generative capabilities of Large Language Models. Methods: We propose SPVR (Syntax-to-Prompt Vulnerability Repair), a novel framework that collects information from syntax trees, and generates corresponding prompts. Our method consists of three steps: rule design, prompt generation, and patch generation. In the rule design step, our method parses code patches and designs rules to extract relevant contextual information. These rules aid in identifying vulnerability-related issues. In the prompt generation step, our method extracts information from vulnerable code with pre-defined rules, automatically converting them into prompts. We also incorporate the description of CWE (Common Weakness Enumeration) as known information into the prompts. Finally, in the patch generation step, this prompt will serve as input to any conversational LLM to obtain code patches. Results: Extensive experiments validate that our method achieves excellent results in assisting LLMs to fix vulnerabilities accurately. We utilize multiple Large Language Models to validate the effectiveness of our work, repairing 143 of 547 vulnerable code using ChatGPT-4. We conducted a comparison of our approach against several existing vulnerability repair approaches (including fine-tuning-based and prompt-based), across multiple metrics.

cs.SE↗

arXiv2025

Automated Prompt Generation for Code Intelligence: An Empirical study and Experience in WeChat

Large Code Models (LCMs) show potential in code intelligence, but their effectiveness is greatly influenced by prompt quality. Current prompt design is mostly manual, which is time-consuming and highly dependent on specific LCMs and tasks. While automated prompt generation (APG) exists in NLP, it is underexplored for code intelligence. This creates a gap, as automating the prompt process is essential for developers facing diverse tasks and black-box LCMs. To mitigate this, we empirically investigate two important parts of APG: Instruction Generation (IG) and Multi-Step Reasoning (MSR). IG provides a task-related description to instruct LCMs, while MSR guides them to produce logical steps before the final answer. We evaluate widely-used APG methods for each part on four open-source LCMs and three code intelligence tasks: code translation (PL-PL), code summarization (PL-NL), and API recommendation (NL-PL).Experimental results indicate that both IG and MSR dramatically enhance performance compared to basic prompts. Based on these results, we propose a novel APG approach combining the best methods of the two parts. Experiments show our approach achieves average improvements of 28.38% in CodeBLEU (code translation), 58.11% in ROUGE-L (code summarization), and 84.53% in SuccessRate@1 (API recommendation) over basic prompts. To validate its effectiveness in an industrial scenario, we evaluate our approach on WeChat-Bench, a proprietary dataset, achieving an average MRR improvement of 148.89% for API recommendation.

cs.SE↗